The Incomplete Grammar approach to the development of a Strong-AI based ICALL system

The development of our ICALL system is supported by the Incomplete Grammar approach, which allows us to model the learner with a cognitively based system of hypotheses held by the Tutor module. An Incomplete Grammar model of a learner’s language performance includes Constituent Productions PC, Constituent Production Errors PCE , Lexical Productions PCL and Lexical Production Errors PLE in the learner model, but is incomplete in the target language space. The generation of the set of hypotheses of mal-rules (elements of PCE, and PLE) when language errors are detected is examined from a computational perspective. Dynamic evaluation and ranking of mal-rules is a challenge. We report on approaches to this problem. Strategies for further development are explored.